EP3336571A1 - Attribution d'empreintes digitales mr sur la base de réseaux neuronaux - Google Patents

Attribution d'empreintes digitales mr sur la base de réseaux neuronaux Download PDF

Info

Publication number
EP3336571A1
EP3336571A1 EP17164986.6A EP17164986A EP3336571A1 EP 3336571 A1 EP3336571 A1 EP 3336571A1 EP 17164986 A EP17164986 A EP 17164986A EP 3336571 A1 EP3336571 A1 EP 3336571A1
Authority
EP
European Patent Office
Prior art keywords
fingerprint
neural network
voxel
parameters
fingerprints
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Withdrawn
Application number
EP17164986.6A
Other languages
German (de)
English (en)
Inventor
Elisabeth Hoppe
Andreas Maier
Josef Pfeuffer
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Siemens Healthcare GmbH
Original Assignee
Friedrich Alexander Univeritaet Erlangen Nuernberg FAU
Siemens Healthcare GmbH
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Friedrich Alexander Univeritaet Erlangen Nuernberg FAU, Siemens Healthcare GmbH filed Critical Friedrich Alexander Univeritaet Erlangen Nuernberg FAU
Priority to EP17164986.6A priority Critical patent/EP3336571A1/fr
Priority to US15/945,964 priority patent/US10698055B2/en
Publication of EP3336571A1 publication Critical patent/EP3336571A1/fr
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/50—NMR imaging systems based on the determination of relaxation times, e.g. T1 measurement by IR sequences; T2 measurement by multiple-echo sequences
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/4828—Resolving the MR signals of different chemical species, e.g. water-fat imaging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
    • G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
    • G01R33/5602—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution by filtering or weighting based on different relaxation times within the sample, e.g. T1 weighting using an inversion pulse
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
    • G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
    • G01R33/5608—Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
    • G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
    • G01R33/561—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution by reduction of the scanning time, i.e. fast acquiring systems, e.g. using echo-planar pulse sequences
    • G01R33/5613—Generating steady state signals, e.g. low flip angle sequences [FLASH]
    • G01R33/5614—Generating steady state signals, e.g. low flip angle sequences [FLASH] using a fully balanced steady-state free precession [bSSFP] pulse sequence, e.g. trueFISP
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H30/00—ICT specially adapted for the handling or processing of medical images
    • G16H30/40—ICT specially adapted for the handling or processing of medical images for processing medical images, e.g. editing
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
    • G01R33/00—Arrangements or instruments for measuring magnetic variables
    • G01R33/20—Arrangements or instruments for measuring magnetic variables involving magnetic resonance
    • G01R33/44—Arrangements or instruments for measuring magnetic variables involving magnetic resonance using nuclear magnetic resonance [NMR]
    • G01R33/48—NMR imaging systems
    • G01R33/54—Signal processing systems, e.g. using pulse sequences ; Generation or control of pulse sequences; Operator console
    • G01R33/56—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution
    • G01R33/561—Image enhancement or correction, e.g. subtraction or averaging techniques, e.g. improvement of signal-to-noise ratio and resolution by reduction of the scanning time, i.e. fast acquiring systems, e.g. using echo-planar pulse sequences
    • G01R33/5613—Generating steady state signals, e.g. low flip angle sequences [FLASH]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks

Definitions

  • the present invention relates to a method for determining MR (magnetic resonance) parameters on the basis of MR fingerprints.
  • MR parameters are important for magnetic resonance techniques, such as magnetic resonance imaging, an imaging technique for imaging body tissue.
  • Some atomic nuclei of the tissue to be examined have an intrinsic angular momentum (nuclear spin) and are therefore magnetic. These nuclei generate a longitudinal magnetization in the direction of the static field after the application of a static magnetic field.
  • a short-term applied high-frequency alternating field in the radio frequency range (RF) the longitudinal magnetization can be deflected from the direction of the static field, so partially or completely convert into a transverse magnetization, whereby an electrical voltage is induced in a receiver coil.
  • the transverse magnetization decreases, the nuclear spins thus align themselves again parallel to the static magnetic field. This so-called relaxation occurs with a characteristic cooldown.
  • Different types of tissue differ in their MR parameters or in their MR parameter combinations, for example in the longitudinal relaxation time T1 and / or in the transverse relaxation time T2.
  • MRF MR fingerprinting
  • the method for obtaining these time series is, for example in Jiang Y, Ma D, N, Gulani V, Griswold MA, "MR Fingerprinting Using Fast Imaging with Steady State Precision (FISP) with Spiral Readout", Magnetic Resonance in Medicine 74: 1621-1631 (2015 ).
  • the MR signals are measured with an MRF-FISP pulse sequence in which an adiabatic inversion pulse is followed by a series of FISP images.
  • the MRF-FISP pulse sequence uses a sinusoidal distribution of adjustment angles and repetition times in a Perlin noise pattern.
  • An interleave of a variable density spiral trajectory is used in each iteration. The spiral trajectory is zero moment compensated.
  • the measured time series i. the MR fingerprint of the tissue
  • a previously simulated MR fingerprint from a dictionary by means of pattern recognition (pattern recognition) (matching).
  • the MR fingerprint directory is previously constructed and includes the simulated fingerprints for possible combinations of all MR parameters.
  • the directory is generated using the Bloch equations and the same acquisition parameters as for the measurement, such as a distribution of flip angles, a phase of the RF pulses, or repetition times.
  • Matching selects the simulated time series from the directory that best matches the measured time series. Since the T1 and T2 relaxation times are known from the directory for the simulated time series, These parameters can also be assigned to the tissue from the corresponding voxel.
  • the grouping is performed by first selecting an initial signal S 0 .
  • the start signal S 0 is compared with all other time series of the directory.
  • the M / N signals most correlated to the initial signal are used to generate the first group.
  • a new signal S 1 is generated, ie an average signal of the first group, which from now on best represents the entries in the first group.
  • a smaller group level of principal component analysis (PCA) is calculated using singular value decomposition. These steps are repeated until all directory elements are distributed.
  • the matching of a detected time series with this compressed directory is performed by matching the time series against all representative group signals S 1 ,..., S N.
  • the groups with the highest matches are selected.
  • the remaining groups are then evaluated by a PCA projection.
  • the best result and its parameters are then selected for the voxel.
  • the object is achieved by a method for determining MR parameters, comprising the steps of detecting an MR fingerprint of a voxel by means of a pulse sequence; inputting the MR fingerprint into the input layer of a trained neural network; and outputting at least one MR parameter to the MR fingerprint at the output layer of the trained neural network.
  • the technical advantage is achieved that a speed of quantification is increased and no directory for the step of quantifying the MR parameters is needed, the storage space requirement grows with the possible parameter combinations contained therein.
  • the directory is implicitly represented by the trained neural network, but - compared to the directory - it has a constant memory size.
  • MR imaging different high frequency signals are collected over a period of time to identify signal evolution for the volume.
  • a spatial resolution (for the assignment of the detected signals to a voxel) is achieved by means of linearly spatially dependent gradient fields.
  • tissue e.g., tumor tissue
  • prior art pattern characterization techniques often use pattern recognition patterns in the signal timing.
  • MRF Magnetic resonance fingerprinting
  • the processing after the MR measurement may include a pattern recognition algorithm to match the collected fingerprints with predicted signal developments from a predefined dictionary, which requires a considerable amount of computation and memory.
  • Neural networks are already used in the field of MR technology. For example, in the patent application US 14 / 682,220 described to use a neural network for generating pulse sequences, but not for classification of measured MR fingerprints.
  • the pulse sequence is an MRF-FISP pulse sequence. This, for example, achieves the technical advantage of being able to obtain MR fingerprints in an efficient manner.
  • the at least one MR parameter comprises a T1 relaxation time and / or a T2 relaxation time.
  • a plurality of MR fingerprints of adjacent voxels are input into the input layer and at least one common MR parameter for the plurality of MR fingerprints is output at the output layer.
  • a single MR fingerprint of a voxel is entered into the input layer and a plurality of MR parameters for adjacent voxels are output at the output layer.
  • a contrast between two MR fingerprints is evaluated by means of the neural network.
  • the technical advantage is achieved that a recording sequence can be changed so that the contrast becomes maximum.
  • the method is used for a plurality of adjacent voxels.
  • the technical Advantage achieved that maps and images of the corresponding MR parameters can be determined.
  • the object is achieved by a computer program having program sections for carrying out the method steps according to the first aspect when the computer program is executed on a computer.
  • the object is achieved by an electronically readable data memory with a computer program according to the second aspect.
  • the object is achieved by a neural network having an input layer for inputting an MR fingerprint of a voxel; multiple hidden layers for processing the MR fingerprint; and an output layer for outputting at least one MR parameter to the MR fingerprint of the voxel.
  • the object is achieved by a magnetic resonance apparatus for determining MR parameters, having a detection unit for detecting an MR fingerprint of a voxel by means of a pulse sequence; and a determining unit for determining at least one MR parameter by inputting the MR fingerprint into the input layer of a trained neural network; and outputting the at least one MR parameter to the MR fingerprint of the voxel at the output layer of the trained neural network.
  • the determination unit may provide a neural network according to the fourth aspect that is trained.
  • the magnetic resonance apparatus is designed to determine MR parameters of a large number of adjacent voxels.
  • the technical advantage is achieved that maps and images of the corresponding MR parameters can be determined.
  • Fig. 1 Figure 12 shows a schematic view of a neural network 100.
  • Deep learning refers to a range of machine learning and a class of artificial neural network optimization methods 100 that have numerous interlayers 101-H between an input layer 101-1 and an output layer 101-0 and have a large internal structure.
  • the net 100 enables a stable learning success even with numerous intermediate layers 101-H.
  • the network 100 may be described as an artificial neural network (ANN) 100 with multiple hidden layer (s) 101-H.
  • the layers 101-H are used to learn abstract features from the input data.
  • the hierarchical representation first learns characteristics at lower levels and then features at higher levels.
  • the artificial neural network 100 is based on information processing of the brain. It comprises a network of calculation units called nodes 103.
  • the nodes 103 are interconnected and comparable to the neurons and axons in the human brain.
  • Each node 103 receives input data (either as actual input data or as output data of another node 103), performs a calculation based on that input data, and forwards the result to connected nodes.
  • the connections between the nodes 103 have weights that define how strong the connection between the respective nodes 103 is.
  • Such a network 100 may include multiple layers 101, such as convolution layers, pooling layers, and / or fully-connected layers.
  • convolution layer not all neurons are connected to the neurons of the previous layer, but filter cores are trained. With the filter cores and the input data, a convolution is performed and the result of the convolution is forwarded to the subsequent activation function.
  • pooling layer With the pooling layer, the resolution is reduced and, for example, the most relevant signals of an environment (for example 2 * 2) are retained (max-pooling).
  • the fully connected layer all output neurons of the previous layer are connected to all the neurons of the fully connected layer.
  • the output value is passed to an activation function.
  • the input layer 101-1 with the input nodes In 1 ,..., In n accepts the data, while the output layer 101-O with the output nodes On 1 , ..., On n as the last layer holds the final output of the network 100.
  • the output layer 101-O with the output nodes On 1 , ..., On n as the last layer holds the final output of the network 100.
  • several hidden layers 101-H may be arranged with the nodes n 11 , ..., n mf .
  • an artificial neural network 100 Before an artificial neural network 100 can be used to calculate output values from input data, it must be trained to learn the weights of the connections between the nodes 103 of the layers 101 and the values in the filters of the convolutional layers.
  • training data is passed through the network 100 and the results are compared to expected basic truth values.
  • a well-known learning algorithm is backpropagation. Thereafter, the error between the results and the expected values is calculated, and the gradient of the error function is used to iteratively change the weights in the artificial neural network 100 and minimize the error.
  • a deep learning method based on the artificial neural network 100 is used instead.
  • the artificial neural network 100 is used to determine the characteristics of an MR time series, i. an MR fingerprint, to learn.
  • the network 100 is then used to determine the quantitative MR parameters, such as the T1 and / or T2 relaxation times, directly from the input MR fingerprint of a voxel.
  • step S101 first, the MR fingerprint 105 of a voxel is detected by means of a pulse sequence.
  • the fingerprint is formed, for example, by plotting the normalized signal intensity I against the number of data points n.
  • the data points represent the time course of the MR signal when using the pulse sequence.
  • step S102 the MR fingerprint 105 of the voxel is input to the input layer 101-1 of the trained neural network 100.
  • the input layer 101-1 comprises a number of n input nodes In 1 , ..., In n .
  • MR fingerprint 105 is passed through the trained network 100.
  • step S103 at least one MR parameter, eg T1, T2, of the voxel 107 to the MR fingerprint 105 at the output layer 101-O of the neural network 100.
  • Fig. 2 10 shows a block diagram for teaching the neural network 100.
  • an artificial neural network 100 for a regression problem is designed and taught.
  • the outputs of the network 100 are continuous numbers, while in a classification problem the output is the probability of one or more probabilities for each of several classes.
  • This network 100 learns the features from the MR time series, i. MR fingerprints, different tissues, such as different T1 and T2 combinations, using a training process and outputs the associated quantitative MR parameters, such as T1 or T2 relaxation times.
  • the time series of different T1 and T2 combinations differ.
  • Network 100 includes convolutional layers, fully connected layers, and non-linear activations.
  • the input to the neural network 100 is a measured time series of the normalized signal intensity of a voxel, i. the MR fingerprint of the voxel, and the output of the neural network 100 are the quantitative MR parameters.
  • the training of the artificial neural network 100 is performed with simulated MR time series, i. MR fingerprints, as input and the associated MR parameters with which the simulation was performed, performed as ground truth values.
  • the data is divided into training, validation and test records.
  • Validation records are used to improve the training results by modifying network 100 hyperparameters.
  • the hyperparameters include, for example, the number of hidden layers in the network, a learning rate, a size and number of filter cores, an optimization method and / or a number of neurons per layer.
  • the basic truth values come, for example, from the application of the described matching methods to the measured signals or are already existing basic truth values, such as in the case of a NIST phantom.
  • step S201 the weights of the neural network 100 are initialized.
  • a normal initialization with random numbers from a Gaussian distribution in a training or a fine-tuning of a stored model for example, a normal initialization with random numbers from a Gaussian distribution in a training or a fine-tuning of a stored model.
  • step S202 input data 201 including, for example, simulated MR fingerprints 105-S and measured MR fingerprints 105-M are supplied.
  • the input data 201 is passed through the network in the forward direction as training data sets.
  • the simulated MR fingerprints 105-S are used.
  • the measured MR fingerprints 105-M are used.
  • step S203 the results are compared with the basic truth values from the directory.
  • the MR parameters associated with the simulated MR fingerprints 105-S are used.
  • Fine tuning uses the phantom parameters of the measured MR fingerprints.
  • step S204 the errors between the corresponding output of the neural network 100 and the basic truth values are calculated.
  • step S205 it is checked whether the predetermined number of iterations has been reached or the error is below a predetermined one Limit lies. If so, the model of the neural network is stored in step S206.
  • step S207 the weights of the nodes 103 in the network 100 are updated by means of backpropagation and a gradient of the error function.
  • Fig. 3 shows a block diagram for a resolution magnification.
  • step S301 an MR fingerprint 105 of a single voxel 107 is input to the neural network 100.
  • step S302 z MR parameters for a plurality of voxels 107,..., 107-9 are subsequently output by the network 100.
  • the network 100 can be trained to enhance a partial volume effect by upsampling the resolution of the MR image or the quantitative maps (super-resolution). This can be accomplished by training the network 100 to output, for example, a 3x3 environment of voxels having their own parameter values that are correspondingly smaller in size for an input MR fingerprint 105 of a single voxel 107.
  • Fig. 4 shows a block diagram for consideration of spatial neighbors.
  • the network 100 may be used to also consider the spatial context of a voxel 107.
  • step S401 the MR fingerprints 105 of a central voxel 107-C and the adjacent voxels 107-N are input to the network 100.
  • the MR fingerprint of the central voxel 107-C and its spatial neighbors 107-N is processed.
  • the spatial context can also be used to estimate error probabilities of the calculations.
  • step S402 the MR parameters for the central voxel 107-C are output through the network 100.
  • the MR parameters for its spatial neighbors can also be predicted.
  • Fig. 5 shows a block diagram for generating a recording sequence, measuring time series and training the neural network.
  • a capture instruction is generated, such as a pulse sequence.
  • the taking rule is used to measure the MR fingerprint of the voxel 107.
  • the network 100 is trained with the measured MR fingerprint.
  • the design of the network 100 can with the teaching, ie the neural network of the patent application US 14 / 682,220 be linked.
  • the web of the patent application US 14 / 682,220 can then be used to produce an ideal acquisition sequence that produces the best possible contrast between two MR fingerprints. These fingerprints can be used to train the network 100 to predict MR parameters.
  • the combination of a deep-learning MRI fingerprinting method allows for the direct prediction of quantitative maps, such as T1 and T2 relaxation times, using a trained neural network from the acquired MR fingerprint of the respective voxels.
  • the known step of matching an MR fingerprint is based on matching a measured time series with a list of simulated time series.
  • the method and / or the neural network may be implemented by a digital computer program having program sections for carrying out the method steps when the computer program is executed on a computer.
  • the computer comprises a computer-readable memory for storing the computer program and a processor for executing the computer program.
  • the computer program may in turn be stored on an external data memory from which the computer program can be loaded into the internal data memory of the computer.
  • the external data store is for example a CD-ROM, a USB flash memory or a storage drive on the Internet.
  • Fig. 6 1 schematically shows a magnetic resonance apparatus 600 for determining MR parameters on the basis of MR fingerprints 105.
  • the magnetic resonance apparatus 600 comprises a detection unit 601 for capturing the MR fingerprint 105 of at least one voxel 107 of an examination subject by means of a pulse sequence, for example a patient 605
  • the detection unit 601 includes a magnet for applying a magnetic field and a control circuit for applying RF pulses and gradient fields.
  • the detection unit 601 may further comprise a receiver unit, for example a coil, which detects the magnetization of the excited nuclear spins by currents induced in the coil.
  • the magnetic resonance apparatus 600 includes a determination unit 603 for determining the MR parameter by inputting the measured MR fingerprint 105 into the input layer 101-1 of the trained neural network 100 and outputting the at least one MR parameter to the MR fingerprint 105 of the voxel 107 at the output layer 101-O of the trained neural network 100.
  • a determination unit 603 for determining the MR parameter by inputting the measured MR fingerprint 105 into the input layer 101-1 of the trained neural network 100 and outputting the at least one MR parameter to the MR fingerprint 105 of the voxel 107 at the output layer 101-O of the trained neural network 100.
  • the determining unit 603 may be constituted by a computer module having a digital memory and a processor capable of executing a computer program for determining the MR parameter. However, the determination unit 603 can also be implemented by an appropriately designed electronic circuit.
  • All method steps may be implemented by means suitable for carrying out the respective method step. All functions performed by objective features may be a method step of a method.

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Artificial Intelligence (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Condensed Matter Physics & Semiconductors (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Radiology & Medical Imaging (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Medical Informatics (AREA)
  • Public Health (AREA)
  • Signal Processing (AREA)
  • Primary Health Care (AREA)
  • Epidemiology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Pathology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Image Analysis (AREA)
EP17164986.6A 2017-04-05 2017-04-05 Attribution d'empreintes digitales mr sur la base de réseaux neuronaux Withdrawn EP3336571A1 (fr)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP17164986.6A EP3336571A1 (fr) 2017-04-05 2017-04-05 Attribution d'empreintes digitales mr sur la base de réseaux neuronaux
US15/945,964 US10698055B2 (en) 2017-04-05 2018-04-05 Method, neural network, and magnetic resonance apparatus for assigning magnetic resonance fingerprints

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
EP17164986.6A EP3336571A1 (fr) 2017-04-05 2017-04-05 Attribution d'empreintes digitales mr sur la base de réseaux neuronaux

Publications (1)

Publication Number Publication Date
EP3336571A1 true EP3336571A1 (fr) 2018-06-20

Family

ID=58578827

Family Applications (1)

Application Number Title Priority Date Filing Date
EP17164986.6A Withdrawn EP3336571A1 (fr) 2017-04-05 2017-04-05 Attribution d'empreintes digitales mr sur la base de réseaux neuronaux

Country Status (2)

Country Link
US (1) US10698055B2 (fr)
EP (1) EP3336571A1 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109325985A (zh) * 2018-09-18 2019-02-12 上海联影智能医疗科技有限公司 磁共振图像重建方法、装置和计算机可读存储介质

Families Citing this family (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10634748B2 (en) * 2016-06-22 2020-04-28 Comsats Institute Of Information Technology Graphical processing unit (GPU) implementation of magnetic resonance fingerprinting (MRF)
US10527695B2 (en) * 2016-12-05 2020-01-07 The General Hospital Corporation Systems and methods for efficient magnetic resonance fingerprinting scheduling
WO2019028763A1 (fr) * 2017-08-10 2019-02-14 Beijing Didi Infinity Technology And Development Co., Ltd. Système et procédé d'estimation d'un temps de parcours et d'une distance
US10761167B2 (en) * 2018-05-18 2020-09-01 Case Western Reserve University System and method for generating a magnetic resonance fingerprinting dictionary using semi-supervised learning
US11776171B2 (en) 2018-09-18 2023-10-03 Shanghai United Imaging Intelligence Co., Ltd. Systems and methods for magnetic resonance image reconstruction
US11948676B2 (en) 2018-12-14 2024-04-02 The Board Of Trustees Of The Leland Stanford Junior University Qualitative and quantitative MRI using deep learning
EP3680681A1 (fr) * 2019-01-10 2020-07-15 Siemens Healthcare GmbH Choisir un schéma d'acquisition de prise d'empreintes digitales par résonance magnétique et un dictionnaire correspondant en fonction d'un pré-balayage
CN110400298B (zh) * 2019-07-23 2023-10-31 中山大学 心脏临床指标的检测方法、装置、设备及介质
WO2021034708A1 (fr) * 2019-08-16 2021-02-25 The Board Of Trustees Of The Leland Stanford Junior University Réglage rétrospectif de contraste de tissu mou dans une imagerie par résonance magnétique
GB202004601D0 (en) * 2020-03-30 2020-05-13 Ind Tomography Systems Plc Apparatus and method for determining a characteristic of a material
CN111798439A (zh) * 2020-07-11 2020-10-20 大连东软教育科技集团有限公司 线上线下融合的医疗影像质量判读方法、系统及存储介质
US11740306B2 (en) 2021-11-30 2023-08-29 Shanghai United Imaging Healthcare Co., Ltd. Systems and methods for magnetic resonance T1 mapping

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150302297A1 (en) * 2014-04-19 2015-10-22 Case Western Reserve University Magnetic Resonance Fingerprinting (MRF) Serial Artificial Neural Network (ANN) Sequence Design
US20160061922A1 (en) * 2014-08-29 2016-03-03 Siemens Aktiengesellschaft Method and apparatus for magnetic resonance examination of an examination object

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6975894B2 (en) * 2001-04-12 2005-12-13 Trustees Of The University Of Pennsylvania Digital topological analysis of trabecular bone MR images and prediction of osteoporosis fractures
DE102006001681B4 (de) * 2006-01-12 2008-07-10 Wismüller, Axel, Dipl.-Phys. Dr.med. Verfahren und Vorrichtung zur Darstellung mehrkanaliger Bilddaten
KR101912715B1 (ko) * 2012-11-20 2018-10-29 삼성전자주식회사 방사선이 방출된 위치의 분포를 추정하는 방법 및 장치
JP6211182B2 (ja) * 2013-06-25 2017-10-11 カウンシル オブ サイエンティフィック アンド インダストリアル リサーチ 仮想スクリーニング用の演算による炭素及びプロトンｎｍｒ化学シフトベースのバイナリフィンガープリント
US10143389B2 (en) * 2014-04-22 2018-12-04 Case Western Reserve University Distinguishing diseased tissue from healthy tissue based on tissue component fractions using magnetic resonance fingerprinting (MRF)
US9964616B2 (en) * 2014-05-30 2018-05-08 Case Western Reserve University Fast group matching for magnetic resonance fingerprinting reconstruction

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150302297A1 (en) * 2014-04-19 2015-10-22 Case Western Reserve University Magnetic Resonance Fingerprinting (MRF) Serial Artificial Neural Network (ANN) Sequence Design
US20160061922A1 (en) * 2014-08-29 2016-03-03 Siemens Aktiengesellschaft Method and apparatus for magnetic resonance examination of an examination object

Non-Patent Citations (9)

* Cited by examiner, † Cited by third party
Title
"Fingerprinting Using Fast Imaging with Steady State Precession (FISP) with Spiral Readout", MAGNETIC RESONANCE IN MEDICINE, vol. 74, 2015, pages 1621 - 1631
AMTHOR, THOMAS ET AL.: "Comparison of Different Approaches of Pattern Matching for MR Fingerprinting", INTERNATIONAL SOCIETY FOR MAGNETIC RESONANCE IN MEDICINE, ISMRM, 2030 ADDISON STREET, 7TH FLOOR, BERKELEY, CA 94704 USA, no. 3394, 15 May 2015 (2015-05-15), XP040669070 *
CAULEY SF; SETSOMPOP K; MA D; JIANG Y; YE H; ADALSTEINSSON E; GRISWOLD MA; WALD LL: "Fast Group Matching for MR Fingerprinting Reconstruction", MAGN. RESON. MED., vol. 74, no. 2, August 2015 (2015-08-01), pages 523 - 528, XP055417887, DOI: doi:10.1002/mrm.25439
GOLKOV VLADIMIR ET AL: "q-Space Deep Learning: Twelve-Fold Shorter and Model-Free Diffusion MRI Scans", IEEE TRANSACTIONS ON MEDICAL IMAGING, IEEE SERVICE CENTER, PISCATAWAY, NJ, US, vol. 35, no. 5, 1 May 2016 (2016-05-01), pages 1344 - 1351, XP011607959, ISSN: 0278-0062, [retrieved on 20160429], DOI: 10.1109/TMI.2016.2551324 *
JAIN A K ET AL: "STATISTICAL PATTERN RECOGNITION: A REVIEW", IEEETRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, IEEE COMPUTER SOCIETY, USA, vol. 22, no. 1, 1 January 2000 (2000-01-01), pages 4 - 37, XP000936788, ISSN: 0162-8828, DOI: 10.1109/34.824819 *
JUAN M. GARCIA-GOMEZ ET AL: "Benign /malignant classifier of soft tissue tumors using MR imaging", MAGNETIC RESONANCE MATERIALS IN PHYSICS, BIOLOGY AND MEDICINE, vol. 16, no. 4, 1 March 2004 (2004-03-01), DE, GB, pages 194 - 201, XP055727556, ISSN: 0968-5243, DOI: 10.1007/s10334-003-0023-7 *
MA D; GULANI V; SEIBERLICH N, LIU K; SUNSHINE JL; DUERK JL; GRISWOLD MA: "Magnetic Resonance Fingerprinting", NATURE, vol. 495, 2013, pages 187 - 192, XP055183037, DOI: doi:10.1038/nature11971
WANG Z; ZHANG Q; YUAN J; WANG X: "Biomedical Imaging (ISBI", 2014, BIOMEDICAL IMAGING (ISBI), 2014 IEEE 11TH INTERNATIONAL SYMPOSIUM ON, article "MRF Denoising with Compressed Sensing and Adaptive Filtering"
YUN JIANG ET AL: "MR fingerprinting using fast imaging with steady state precession (FISP) with spiral readout : MR Fingerprinting with FISP", MAGNETIC RESONANCE IN MEDICINE., vol. 74, no. 6, 9 December 2014 (2014-12-09), US, pages 1621 - 1631, XP055417885, ISSN: 0740-3194, DOI: 10.1002/mrm.25559 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109325985A (zh) * 2018-09-18 2019-02-12 上海联影智能医疗科技有限公司 磁共振图像重建方法、装置和计算机可读存储介质
CN109325985B (zh) * 2018-09-18 2020-07-21 上海联影智能医疗科技有限公司 磁共振图像重建方法、装置和计算机可读存储介质

Also Published As

Publication number Publication date
US10698055B2 (en) 2020-06-30
US20180292484A1 (en) 2018-10-11

Similar Documents

Publication Publication Date Title
EP3336571A1 (fr) Attribution d'empreintes digitales mr sur la base de réseaux neuronaux
EP3299833B1 (fr) Génération améliorée des séries chronologiques de points image d'un objet examiné au moyen de la technique de résonance magnétique
DE102013205785B4 (de) Ermittlung einer Magnetresonanzsystem-Ansteuersequenz auf Basis einer reduzierten Anzahl an Feldverteilungskarten
DE102013217651B4 (de) Mehrpunkt Dixon-Technik
DE102007044463A1 (de) Verfahren zur Bestimmung der räumlichen Verteilung von Magnetresonanzsignalen durch mehrdimensionale HF-Anregungspulse
DE102013201814B4 (de) Verfahren zur Magnetresonanz-Bildgebung mit multidimensional ortsselektiven HF-Pulsen in einem Außenbereich
DE102013217650B4 (de) Zwei-Punkt Dixon-Technik
WO2020229243A1 (fr) Procédé de prise d'empreinte numérique par résonance magnétique
DE102015207590A1 (de) Verfahren zu einer Bewegungskompensation während einer Magnetresonanz-Bildgebung
DE112017005080T5 (de) Direkte messung des b0-ausserresonanzfelds während magnetresonanz-fingerabdruckerzeugung
DE102015218106A1 (de) Verfahren zu einer Bewegungskorrektur von Magnetresonanz-Messdaten
DE102015220078B4 (de) Erstellung von reduzierten Signalverläufen für eine Magnetresonanz-Fingerprinting-Untersuchung eines Untersuchungsobjekts
DE102012204625B4 (de) Bestimmen eines Gesamtparameters einer Pulssequenz anhand einer Baumstruktur
DE102020210775A1 (de) Magnetresonanztomographie-Rekonstruktion durch Verwenden von maschinellem Lernen und Bewegungskompensation
DE102016200629A1 (de) Verfahren zur Magnetresonanz-Bildgebung
DE102016207638A1 (de) Multischicht Matgentresonanz Fingerprinting
DE102014212943B4 (de) Magnetresonanz-Bildgebung unter Berücksichtigung von unterschiedlichen Frequenzkodiermustern
DE102018209584A1 (de) Magnetresonanz-Fingerprinting-Verfahren
DE102010041801B4 (de) MR-Einzelpunkt-Bildgebung mit verkürzter Echozeit
EP3828576B1 (fr) Procédé de génération d'au moins un ensemble de données d'image et d'un ensemble de données d'image de référence, support de données, produit programme informatique ainsi qu'installation de résonance magnétique
EP3828581B1 (fr) Procédé de génération d'au moins un ensemble de données d'image et d'un ensemble de données d'image de référence, produit programme informatique ainsi qu'installation de résonance magnétique
DE102014222495A1 (de) Korrektur von Magnetresonanz-Bildaufnahmen mit Hilfe konvertierter Feldkarten
DE102018213326A1 (de) Mehrpunkt-Magnetresonanz-Bildgebung
EP3336572A1 (fr) Détermination de paramètres de diffusion utilisant la prise d'empreinte par résonance magnétique nucléaire
DE102016200550A1 (de) Verfahren zu einer Analyse einer Magnetresonanz-Sequenz

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20180419

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

AX Request for extension of the european patent

Extension state: BA ME

RAP1 Party data changed (applicant data changed or rights of an application transferred)

Owner name: SIEMENS HEALTHCARE GMBH

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: EXAMINATION IS IN PROGRESS

17Q First examination report despatched

Effective date: 20200910

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION IS DEEMED TO BE WITHDRAWN

18D Application deemed to be withdrawn

Effective date: 20210121